Swami NatarajanJune 13, 2015 RIT Software Engineering Project Management

Swami NatarajanJune 13, 2015 RIT Software Engineering Project Management Metrics Cycletime Productivity Staffing Requirements volatility Reuse metrics Activity progress measurement Estimation accuracy

Swami NatarajanJune 13, 2015 RIT Software Engineering Cycletime Time from requirements to release (one cycle) Constant pressure in the corporate world to improve cycletime –Improves time-to-market Getting to market ahead of competition has big impact on market share, profits –Correlates heavily with cost –Reduces gap between market survey and actual market by release time Also important for custom solutions –Getting deliverable earlier to customer saves them money (increases value of deliverable)

Swami NatarajanJune 13, 2015 RIT Software Engineering Impact of time-to-market Product Selling Price Early arrival’s costs Late arrival’s costs Does not show market share impacts! Premium Products of early and late arrival both mature over time, reducing costs, but early arrival has higher maturity at any given time $ time

Swami NatarajanJune 13, 2015 RIT Software Engineering Practices for cycletime reduction Incremental development (  agile development) –Quicker release cycle makes it easier to get new features into product quickly –Break up 12-month cycle into 4 cycles of 4 months each! (yes, that makes sense!) Use of tools and technologies that improve productivity More concurrent engineering (increases coordination needs) Planning and risk management to avoid holdups –Rightsizing teams to minimize development cycletime Avoid building from scratch: use existing libraries and products where possible –Invest in developing libraries and domain architectures Streamlining development through checklists, templates, workflow

Swami NatarajanJune 13, 2015 RIT Software Engineering Measuring cycletime Basically simple: project start date, end date Project cycletime vs. development cycletime –Development time: requirements-to-release –May expend a lot of time before requirements phase! Issue: what about holdups “beyond one’s control”? –May have a concept of “stopping cycletime clock” –Shows the need for proper operational definitions –Note the possibility of superior practices that avoid holdups Measurements & metrics can impact which practices encouraged!

Swami NatarajanJune 13, 2015 RIT Software Engineering Cycletime Metrics Challenging to create metric for cycletime –Are projects really “comparable”? Different features, different complexity Customers may or may not be willing to pay for speed –Avoid encouraging “bad practices” e.g. unreasonably small increments Release must provide “significant value” to customer “Bucket” concept –Group together “broadly similar” projects and measure Hard to get enough projects for statistical significance More important to compare with competitor cycletimes Focus on “constant improvement”

Swami NatarajanJune 13, 2015 RIT Software Engineering Productivity Objective: Measure effectiveness of organizational practices in getting work done –Measuring individual productivity is a no-no Extremely prone to abuses, creates pressures Impacts teaming, co-operation: “credit-grabbing” Hard to balance with quality Counter-productive in the longer term Measure: size of deliverable / effort expended –Size of deliverable != volume of work (KLOC) Credit for effective reuse / choosing good platforms

Swami NatarajanJune 13, 2015 RIT Software Engineering Productivity Metrics Function-points/staff-month –Better than KLOC / staff-month Avoids problems related to “density of code” Challenges in productivity comparisons –Accounting for complexity (compare only with same domain) But still, not all function points created equal! –Assigning proper value for tools / technology / platform usage “Size of deliverable” gives too much credit (what about added cost?) “Actual work done” gives too little –Impact of other factors Requirements volatility, staff profile, nature of work (fresh / legacy), tough product quality requirements, development infrastructure, time overheads … Interpret with extreme caution! –Minefield, easy to overemphasize because it is so “bottom line”

Swami NatarajanJune 13, 2015 RIT Software Engineering Using Productivity Numbers Trend information may add value –Indicate whether there is constant improvement of practices Comparison with competitors, industry average –OK measure of overall effectiveness –Beware of differences in measurements, reporting Useful to evaluate technologies and practices Excellent complementary metric –Improvements in other numbers should mostly show up in productivity e.g. COQ/COPQ balancing, fault injection

Swami NatarajanJune 13, 2015 RIT Software Engineering Staffing Curves showing planned & actual staffing for each month –Gaps would indicate potential schedule impacts –Significant increases in planned staffing must be accompanied by training/induction plans May include turnover rates –People moving out, people added –High turnover will impact productivity, schedule Limitation: shows raw numbers, not skill level Metrics –% staffing (actual/planned) –% turnover

Swami NatarajanJune 13, 2015 RIT Software Engineering Staffing chart Planned Actual Lost Added Time Number of people Month1 Month2 Month3 Month4 Month5

Swami NatarajanJune 13, 2015 RIT Software Engineering Requirements volatility Month-by-month percentage change in requirements –Based on either use cases or numbered requirements –Includes added/deleted/changed requirements High requirements volatility impacts schedule, fault injection, productivity –Can use “control line” e.g. 10% - requirements change more than this triggers risk mitigation (impact analysis / replanning) If using tools to manage requirements, relatively easy to generate requirements volatility metrics Limitation: does not show severity/impact of changes

Swami NatarajanJune 13, 2015 RIT Software Engineering Reuse Metrics % of reused code –Hard to define how much to count as reused code “Scavenged code” (cut-paste) is least valuable Libraries better – should we give full credit for each use? Using COTS (commercial-off-the-shelf) software better e.g. don’t write your own OS – how do you count this? Domain engineering – creating standard product architectures and avoiding developing afresh from scratch best – should we give full credit for each use? Common practice: –Measure libraries and/or scavenged code –Can add notes about use of COTS and/or domain architectures and components Note that the end goal is productivity, not reuse

Swami NatarajanJune 13, 2015 RIT Software Engineering Progress Objective: Measure progress against plan –Avoid situation where lateness realized just prior to release Practices –Define milestones 2-3 weeks apart –Measure planned vs. actual completion dates –If 2 weeks or more behind schedule, replan Re-negotiate fresh delivery dates with customer Metric –Chart of planned and actual completion dates –% slippage: (actual – planned ) / planned completion time

Swami NatarajanJune 13, 2015 RIT Software Engineering Progress: Milestone chart MilestonePlannedActual Initial requirements14-Mar13-Mar Prototype4-Apr6-Apr Requirements baselined12-Apr Initial design23-Apr28-Apr V1 code complete8-May24-May (replan) Integration done12-May 28-May 29-May Release 11-June 12-June

Swami NatarajanJune 13, 2015 RIT Software Engineering Progress: Earned Value Charts A superior way to measure progress For each activity, define an “earned value” – some number of points –Assign more earned value if more effort needed Track actual earned value –Total points earned for all completed activities –Irrespective of actual effort expended –May add another curve that shows actual effort expended Plot planned vs. actual earned value against time –Shows % completion of project very clearly Applicable to agile development too!

Swami NatarajanJune 13, 2015 RIT Software Engineering Earned value chart: Example From Lethbridge & Laganiere, “Object-oriented software engineering”

Swami NatarajanJune 13, 2015 RIT Software Engineering Gantt charts From Lethbridge & Laganiere, “Object-oriented software engineering”

Swami NatarajanJune 13, 2015 RIT Software Engineering Estimation Accuracy (Actual effort – Estimated effort )/ Estimated effort Typically around 20% (i.e. 20% underestimate) for “good” organizations –Note that this often doesn’t translate to 20% slippage – either replanning or overtime work! –If maturity is low, maybe 50% - 100% or even more! Can track estimation accuracy for initial estimates as well as for “final” estimates –Initial estimates may be prior to understanding requirements Correlate with requirements volatility to get better picture Limitation: “Work expands to fill time available” –Hard to detect overestimates

Swami NatarajanJune 13, 2015 RIT Software Engineering Summary Can track a variety of metrics that reflect various project management concerns Used to detect likelihood of various problems –slippage, productivity loss, need for training Correlate multiple curves to assess health of project –Typically all these curves on one big chart Each metric vulnerable to abuse –Need to be careful how we use them